AI Journalism & Credibility

Motivation

Machine authorship, where software or programming is the primary creator of news articles (Danzon-Chambaud, 2021; Lewis, 2019), is a topic of ongoing debate. Scholars hold differing views on how machine authorship impacts news audiences’ perceptions, with results varying significantly across studies. This inconsistency may stem from individual differences among study participants, suggesting that future research might continue to yield varied outcomes rather than a unified consensus. From a theoretical perspective, the MAIN model (Sundar, 2008) suggests that people’s pre-existing beliefs about machines, shaped by their knowledge and experience, influence their perceptions. When engaging with machine-generated content, these beliefs often activate “machine heuristics”—mental shortcuts triggered by machine interface cues.

Study design

We used the GPT-4 Turbo model to create AI-authored news articles. The model’s input consisted of published articles, with the goal of generating summaries in short paragraphs for a general audience. Specific guidelines instructed the model to refer to the original authors in the third person (e.g., “the authors” or “the researchers”) and to ensure accessibility for a wider public audience. All provided quotes were required to be incorporated directly into the output.

These AI-generated articles were then compared to the original scientific articles written by journalists based on the same published papers.

Hypothesized Framework